The Strengths and Difficulties Questionnaire as a

Predictor of Parent-Reported Diagnosis of Autism

Spectrum Disorder and Attention Deficit Hyperactivity
Disorder
Ginny Russell1*, Lauren R. Rodgers2, Tamsin Ford1
1 Institute of Health Services Research, University of Exeter Medical School, Exeter, United Kingdom, 2 NIHR CLAHRC for the South West Peninsula (PenCLAHRC), University
of Exeter Medical School, Exeter, United Kingdom

Abstract
The Strengths and Difficulties Questionnaire (SDQ) is widely used as an international standardised instrument measuring
child behaviour. The primary aim of our study was to examine whether behavioral symptoms measured by SDQ were
elevated among children with autism spectrum disorder (ASD) and attention deficit hyperactivity disorder (ADHD) relative
to the rest of the population, and to examine the predictive value of the SDQ for outcome of parent-reported clinical
diagnosis of ASD/ADHD. A secondary aim was to examine the extent of overlap in symptoms between children diagnosed
with these two disorders, as measured by the SDQ subscales. A cross-sectional secondary analysis of data from the
Millennium Birth Cohort (n = 19,519), was conducted. Data were weighted to be representative of the UK population as a
whole. ADHD or ASD identified by a medical doctor or health professional were reported by parents in 2008 and this was
the case definition of diagnosis; (ADHD n = 173, ASD n = 209, excluding twins and triplets). Study childrens ages ranged
from 6.38.2 years; (mean 7.2 years). Logistic regression was used to examine the association between the parent-reported
clinical diagnosis of ASD/ADHD and teacher and parent-reported SDQ subscales. All SDQ subscales were strongly associated
with both ASD and ADHD. There was substantial co-occurrence of behavioral difficulties between children diagnosed with
ASD and those diagnosed with ADHD. After adjustment for other subscales, the final model for ADHD, contained
hyperactivity/inattention and impact symptoms only and had a sensitivity of 91% and specificity of 90%; (AUC) = 0.94 (95%
CI, 0.900.97). The final model for ASD was composed of all subscales except the peer problems scales, indicating of the
complexity of behavioural difficulties that may accompany ASD. A threshold of 0.03 produced model sensitivity and
specificity of 79% and 93% respectively; AUC = 0.90 (95% CI, 0.860.95). The results support changes to DSM-5 removing
exclusivity clauses.

Citation: Russell G, Rodgers LR, Ford T (2013) The Strengths and Difficulties Questionnaire as a Predictor of Parent-Reported Diagnosis of Autism Spectrum
Disorder and Attention Deficit Hyperactivity Disorder. PLoS ONE 8(12): e80247. doi:10.1371/journal.pone.0080247
Editor: Atsushi Senju, Birkbeck, University of London, United Kingdom
Received May 23, 2013; Accepted October 1, 2013; Published December 3, 2013
Copyright: 2013 Russell et al. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits
unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Funding: Funding was provided by the ESRC as part of the Secondary Data Analysis Initiative, Grant Number ES/K003356/1. The authors also acknowledge
funding from the National Institute for Health Research (NIHR) Collaboration for Leadership in Applied Health Research and Care (CLAHRC) for the South West
Peninsula. The views and opinions expressed in this paper are those of the authors and not necessarily those of the NHS, the NIHR or the Department of Health.
The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Competing Interests: The authors have declared that no competing interests exist.
* E-mail: g.russell@ex.ac.uk

Introduction The SDQ has been used in in clinical practice as a screening

The Strengths and Difficulties Questionnaire (SDQ) is a brief
dimensional measure of psychopathology among children aged
416 that has been widely adopted in both research and in
clinical practice [1]. The instrument is composed of 25 items that
ask about behavioral attributes of the child and are combined to
form five subscales (composed of 5 items each). The subscales
measure emotional symptoms, conduct problems, hyperactivity/
inattention, peer relationships, and prosocial behavior. There are
parallel versions of the SDQ that collect the same data from
parents, teachers and young people aged 11 or over. A
supplemental impact subscale measures chronicity, distress,
social impairment, and burden to others, which provides useful
additional information for clinicians and researchers [2].
and/or assessment tool by both school psychologists [3] and
clinicians [2,4,5]. It is also used extensively in research studies
throughout Europe [68] the USA [9,10], Asia [5,1113] and
Africa [13]. To date, the SDQ has received over 3,000 research
citations and this number is growing, particularly as many on-
going longitudinal birth cohorts have used the SDQ for over a
decade as a repeated measure of child behaviour [7,14,15].
Woerner and colleagues [8] reviewed non-European studies
that psychometrically evaluated the SDQ, applied it to screen for
behaviour disorders, or employed its parent-, teacher- or self-rated
versions as research tools. They found experience gained with the
SDQ in other continents has supported European evidence of
good psychometric properties and clinical utility. They note that
worldwide usage of the SDQ is expected to increase in the future,
although reporting by different participants is context-dependent

PLOS ONE | www.plosone.org 1 December 2013 | Volume 8 | Issue 12 | e80247


Figure 1. Box plots for parent report of SDQ subscales across three groups: ASD diagnosis, ADHD diagnosis and neither diagnosis.
Diagnosis: dx Increasing score reflects increased impairment in all sub-scales except prosocial scores which measure strengths.
doi:10.1371/journal.pone.0080247.g001
and this limits the reliability of cross-cultural comparisons [7,16].
Despite these reservations, the SDQ has been successfully used to
make comparisons of child behaviour across age and culture [17].
Various studies have examined the utility of the SDQ as a
screening device in predicting childhood psychiatric cases [1820]
although few have looked at SDQ as a screen for specific disorders.
In a UK community-based sample, multi-informant ratings
[parents, teachers and older children] identified individuals with
specific psychiatric diagnoses [21]. Sensitivity was over 70% for
identifying conduct and hyperactivity disorders, but the instrument
had poor discrimination (,30%) for emotional disorders in this
general population sample. Varying results are most likely due to
the heterogeneity of symptoms of childhood emotional disorders
which have a wide range of symptoms, only some of which are
captured by the five questions about emotional difficulties in the
SDQ. Notably, there are no questions that directly relate to the
triad of difficulties that comprise the autism spectrum, although
indirectly, social skills can be inferred from the prosocial and peer
relationship subscales. In contrast, there are five questions each on
the more homogenous area of difficulties with attention/
hyperactivity.
Goodman and colleagues [21] developed an SDQ algorithm
that combines teacher, parent and child reports, to predict various
disorders, including Probable Hyperactive Disorder (PHD) in
children. The PHD algorithm uses a combination of informants
for SDQ scores on the hyperactivity/inattention and impact
subscales [18,22]. Multiple informants are required because
symptoms must be present across multiple settings if ADHD is
to be diagnosed [23,24]. Ullebo and colleagues [20] tested the
PHD algorithm and found that it had an acceptable sensitivity for
the ADHD combined phenotype. They recommended that
bespoke cut-offs should be developed according to the purpose
of its application to research. Brndbo and colleagues [25]
cautioned against use of the PHD algorithm as a screening
PLOS ONE | www.plosone.org
The Association of SDQ with Diagnosis of ASD/ADHD
instrument for ADHD in the clinic because of the large number of
false positives identified.
According to the International Classification of Diseases (ICD-
10), for a diagnosis of ASD to be made, children must display
impairments in social interactions and communication, as well as
restricted interests and repetitive behaviour [23]. Iizuka and
colleagues [26] examined the co-occurrence of behavioural
symptoms of high-functioning ASD and ADHD using the SDQ
subscales in Japan. Core symptoms of ASD include social and
communication impairments and, as expected, the two SDQ
subscales that measure aspects of social interaction- peer problems
and prosocial behaviour- were associated with ASD in particular.
The study found elevated levels of peer problems and emotional
difficulties, and fewer prosocial behaviours among the ASD group
compared to those children with ADHD, whilst higher levels of
hyperactivity and more conduct problems were reported for
children with ADHD. A large and growing literature has
demonstrated that ADHD symptoms are relatively common
among children and adults with ASD and vice-versa [2733].
Recently, some researchers have queried whether ASD and
ADHD should be considered as different manifestations of one
overarching disorder [33,34]. Currently, the diagnostic criteria for
childhood disorders laid out in ICD-10 contain an exclusivity
clause that does not allow ADHD to be diagnosed if pervasive
developmental disorder (including ASD) is present, although the
exclusivity clause has been dropped in the new version of the
Diagnostic and Statistical Manual of Mental Disorders (DSM-5),
in which ASD is listed as a condition which is commonly comorbid
with ADHD [24].
Given the wide and on-going use of the SDQ in research on
developmental disorders, we sought to clarify the predictive power
of the SDQ subscales in the identification of parent-reported
clinical diagnosis of two specific disorders: ADHD and ASD. The
primary aim of our study was to examine whether all behavioral
symptoms measured by SDQ were elevated in children with ASD
2 December 2013 | Volume 8 | Issue 12 | e80247

Figure 2. Box plots for teacher report of SDQ subscales across three groups: ASD diagnosis, ADHD diagnosis and neither diagnosis.
Diagnosis: dx Increasing score reflects increased impairment in all sub-scales except prosocial scores which measure strengths.
doi:10.1371/journal.pone.0080247.g002
and ADHD relative to the rest of the population, and the utility of
the SDQ as an indicator of these disorders. A secondary aim was
to examine the extent to which symptoms co-occurred in children
diagnosed with ASD or ADHD. We hypothesised hyperactivity/
inattention symptoms would predict clinical diagnosis of ADHD,
and prosocial and peer relationship problems would predict ASD.
This study therefore assesses the utility of the SDQ in identifying
these disorders.
Methods
Ethics Statement
Information was gathered from the sample, the first Millennium
Cohort Study (MCS) survey when children were 9 months old,
and three, five and seven years of age: four sweeps of data
collection [35]. Informed written consent was obtained at each
stage of the study from parents for their participation and the
participation of their child (ren); the MCS ethical review gives
details [36]. Written consent was also obtained for gathering
information from health, education and economic records and to
contact teachers. The data were analyzed anonymously, with
researchers having no access to participant identities. Identities are
protected by the curators of the MCS. Additional ethical approval
for the analysis reported here was granted by the Peninsula
Medical School Ethics committee.
Design
Our study sought to clarify the predictive power of the SDQ
subscales in the identification of parent-reported diagnosis of ASD
and ADHD using logistic regression models. This was compared
to the predictive power of the PHD algorithm already in existence
[21]. A secondary aim was to examine the extent of overlap in
symptoms between children diagnosed with these two disorders, as
measured by the SDQ subscales, in order to inform the debate
about revisions to diagnostic criteria.
PLOS ONE | www.plosone.org
The Association of SDQ with Diagnosis of ASD/ADHD
Sample
The Millennium Cohort Study (MCS) is a UK-representative
birth cohort study that used a disproportionate stratified cluster
sampling design [35,37]. Children born between 1st September
2000 and 11th January 2002 and listed on the Child Benefit
Records were eligible for the study. Child Benefit was a financial
benefit payable to all parents of UK children at this time, with
near universal take up. Data were first collected when children
were 9 months old (1st wave), further data were recorded
concerning the childrens health and development when the
children were 3 years old (2nd wave), 5 years old (3rd wave) and
7 years old (4th wave). Within the total MCS cohort of 19, 519
children, the current study outcomes, ASD and ADHD status,
were recorded for 14, 043 children at wave 4 (over 70%). The
MCS provides appropriate standardised weightings to adjust for
the effect of attrition and oversampling, making these results
representative of the UK population as a whole. Details of
sampling design and weighting calculations are documented in
detail elsewhere [37].
Outcome measures
The case definition of the two conditions was based on
responses to an MCS question duplicated from the US National
Health Interview Survey questionnaire reported in previous
studies [38]. Parents or carers were asked in face-to-face interviews
if a doctor or health professional had identified childhood ADHD
or ASD. Consistent with other studies using these data [39],
families with twins or triplets where other siblings participated
were excluded (252 twins, 11 triplets) as both diagnoses have a
high heritability. Parent-reported ASD and/or ADHD diagnosis
was recorded for 14,043 children in 2008/9 with the wording of
the following questions read out verbatim:
N Has a doctor or health professional ever told you that (sample child) had
attention deficit hyperactivity disorder (ADHD)
3 December 2013 | Volume 8 | Issue 12 | e80247
 The Association of SDQ with Diagnosis of ASD/ADHD

Table 1. Models showing SDQ sub-scales as predictors with ADHD diagnosis as outcome for children from Milenium Cohort at age
7.

Unadjusted Unadjusted Adjusted Adjusted OR Adjusted Final OR

Variable Unadjusted n OR & 95% CI p n & 95% CI p1 Final n & 95% CI3 Final p

Emotion 13082 1.50 (1.40,1.60) ,0.001 8133 1.11 (0.96,1.28) 0.17

Parent
Emotion 8511 1.25 (1.16,1.34) ,0.001 8133 0.89 (0.77,1.04) 0.13
Teacher
Conduct 13111 1.85 (1.72,1.99) ,0.001 8133 1.07 (0.93,1.23) 0.37
Parent
Conduct 8514 1.49 (1.41,1.58) ,0.001 8133 0.93 (0.79,1.09) 0.36
Teacher
Hyper 13061 2.17 (1.95,2.42) ,0.001 8133 1.56 (1.34,1.83) ,0.001 8277 1.56 (1.35,1.80) ,0.001
Parent
Hyper 8512 1.60 (1.48,1.72) ,0.001 8133 1.19 (1.06,1.35) 0.003 8277 1.22 (1.11,1.34) ,0.001
Teacher
Prosocial 13116 0.65(0.59,0.70) ,0.001 8133 0.99 (0.87,1.13) 0.87
Parent2
Prosocial 8510 0.68(0.63,0.72) ,0.001 8133 1.05 (0.92,1.19) 0.46
Teacher2
Peer 13094 1.68 (1.56,1.80) ,0.001 8133 0.93 (0.81,1.07) 0.32
Parent
Peer 8511 1.53 (1.42,1.64) ,0.001 8133 1.09 (0.93,1.28) 0.28
Teacher
Impact 12958 2.17 (1.96,2.41) ,0.001 8133 1.56 (1.37,1.78) ,0.001 8277 1.63 (1.45,1.84) ,0.001
Parent
Impact 8404 2.19 (1.98,2.41) ,0.001 8133 1.12 (0.92,1.36) 0.25
Teacher

1 Adjusted models include subscales significant at 10% levels. 2.For prosocial scores the reciprocal of the odds ratios.
Is presented to fit conceptually with the rest of the model, i.e. greater OR = greater association with ADHD. 3. Model constant 0.0002 on odds ratio scale.
doi:10.1371/journal.pone.0080247.t001

N Has a doctor or health professional ever told you that (sample child) had
autism, Aspergers syndrome or autistic spectrum disorder?
Families at wave 4 whose study children were seven years old,
who responded with positive or negative answers to the above
questions, were included. Families who answered dont know or
refused to answer were excluded from the analysis (n = 30 ASD,
n = 44 ADHD, of these, n = 17 refused/dont know in both
categories). We took this measure to represent a clinical diagnosis
of disorder in line with other studies [38,40,41]. In total, from this
sample, 173 children had reportedly been identified with ADHD
and 209 had a parent-reported ASD diagnosis by age 7. Forty-four
children had a co-morbid diagnosis of ASD and ADHD, and were
retained in both outcome groups.
Independent variables
The SDQ is composed of 25 items that ask about behavioural
attributes of the child and are combined to form five subscales
(composed of 5 items each). The emotional symptoms subscale
contains items that ask about fears, worries, misery, nerves and
somatic symptoms, the conduct problems subscale inquires about
tantrums, obedience, fighting, lying and stealing, and the
hyperactivity/inattention subscale covers restlessness, fidgeting,
concentration, distractibility and impulsivity. The peer relation-
ships subscale items include questions about popularity, victimi-
zation, isolation, friendship and ability to relate to children as
compared to adults, and the prosocial subscale covers consider-
ation of others, ability to share, kindness to younger children, and
helpfulness when other children are distressed and willingness to
volunteer to comfort. For all the subscales except the prosocial
subscale, high scores indicate difficulties. As the prosocial items ask
about the presence of prosocial behaviour, the subscale measures
the strengths of the child in this area, and increasing scores
represent increasingly prosocial behaviour, unlike the other sub-
scales where increasing score represents increasing impairment. In
all cases, answer options for each item are: Not true Somewhat
true or Certainly true, and these are scored 0, 1 or 2, giving a
total score out of a possible 10 for each subscale. A further impact
subscale measures the impact of any difficulties on carers and the
children themselves in terms of chronicity, distress, social
impairment, and burden to others. This is again scored 010
with increasing impact producing a higher score. More details
about the SDQ, the probable hyperactivity disorder (PHD)
algorithm, normative data, background research and how the
subscales are scored are available at the SDQ website (www.
sdqinfo.org).
SDQ scores for each subscale had been taken for the entire
cohort at wave 4 from both parent and teacher informants. Both
were added to models, since clinical identification of the disorders
should be documented as causing impairment across settings (for
example, home and school). Several studies have stressed the need
for information from multiple informants when rating symptoms
of a child psychiatric disorder [42].
Analysis
The ASD, ADHD and general population were compared on
SDQ subscale scores. Box plots were provided for teacher and
parent report of behaviour separately to illustrate how the three

PLOS ONE | www.plosone.org 4 December 2013 | Volume 8 | Issue 12 | e80247


groups (ASD, ADHD and general population) differed in SDQ
scores. Children reported as having both diagnoses (n = 44) were
included in both ASD and ADHD groups.
Logistic regression (LR) established the odds of diagnosis of
ASD/ADHD using SDQ subscales as independent variables.
Parent and teacher ratings of behaviour were treated as separate
covariates. The odds ratios (OR) from the analyses indicate that
the relative increase in odds of being identified with ASD/ADHD
corresponded to a one-point increase in the SDQ subscales. All the
sub-scales bar the prosocial scale measure impairment, therefore
the reciprocal of the odds ratios for the prosocial scores was used to
fit conceptually with the rest of the model. This means that for all
SDQ subscales, an odds ratio greater than 1 represents greater
prediction of diagnosis as childrens difficulties increase. Unad-
justed logistic regression models were fitted in which just one
predictor at a time was included. Multivariable (adjusted) logistic
regression models were then fitted in which predictors significant
at the 10% level in the unadjusted analyses were included as
covariates. Estimates from LR were weighted to take account of
the disproportionate stratified sample of electoral wards and
attrition/non-response by the 4th wave when the study outcomes
were measured, making the sample representative of the UK
population [37]. LR was then used to derive separate models for
ASD and for ADHD respectively, composed of the SDQ subscales
that remained significant at 10% levels after adjustment for other
subscales. Final models were composed of subscales that remained
significantly associated with outcome at 10% levels after adjust-
ment for other behaviours. The sensitivity (percentage of children
with diagnosis correctly identified as such) and specificity
(probability that a test result will be negative when the disease is
not present or true negative rate, expressed as a percentage) of the
final models were examined using Receiver Operating Charac-
teristic (ROC) curves. The area under the curve (AUC) is a
measure of how well the model can identify children with disorder.
The Youden Index [43] is used to calculate the optimal values for
sensitivity and specificity; it determines a threshold that will
maximise the difference between true positive and false positive
rates. For this threshold, the positive predictive value was derived
for each model. In the case of ADHD, the sensitivity and
specificity were compared to the cut-offs for Probable Hyperac-
tivity Disorder algorithm [21].
Results
For 96.7% of families participating, the main respondent on the
outcome measure of ASD or ADHD was the childs mother. At
the birth of the child, mothers had a mean age of 28 years (range
13 to 48 years), and over 99% were resident at home with the
study child all of the time. The mean child age when outcome
measures were taken was 7.2 years (SD = 0.2; range, 6.3 to 8.2).
Figures 1 and 2 illustrate the demographic profile of the sample,
giving descriptive statistics for parent and teacher-rated SDQ
subscales for children with ASD, those with ADHD and those with
neither diagnosis. Clear differences are observed between the
children with neither diagnosis (no dx) and children with ADHD/
ASD. The figures illustrate differences in the distribution of scores
between ASD children and those with ADHD but also substantial
overlap. The inter-rater reliability between parent and teacher
scores was low to medium, values of the weighted kappa coefficient
ranged from 0.24 for the emotional symptoms sub-score (95% CI
0.220.27) to 0.47 for hyperactivity/inattention scores (95% CI
0.4520.47).
As Figure 1 and Figure 2 illustrate, children with ADHD and
ASD diagnoses had substantially more impaired behaviour at age
PLOS ONE | www.plosone.org
The Association of SDQ with Diagnosis of ASD/ADHD
7 than other participating children without either of these
diagnoses on every SDQ sub-scale. Hyperactivity scores were
particularly high in both diagnosed samples compared to those of
the rest of the population. Impact scores were also higher and
prosocial skills were lower in both diagnosed groups according to
both informants.
Table 1 reports the results of LR for the outcome of ADHD.
These results confirm that all subscales were significantly
associated with diagnosis of ADHD, and reflect greater impair-
ment across the range of behaviours measured by the SDQ
instrument. After adjustment for the other SDQ subscales, only
teacher and parent-reported hyperactivity/inattention subscales
and parent-reported impact remained significantly associated with
ADHD diagnosis. A threshold of 0.02 from the model yielded the
optimal sensitivity and specificity values of 91% and 90%
respectively. The positive predictive value (PPV) was low at
12%, which is to be expected in a population based sample
screening for rare disorders comprising young children. The Area
Under the Curve (AUC) = 0.94 (95% CI, 0.900.97) shows the
model is a good fit. The PHD algorithm [21] produces a sensitivity
of just 30%, but a specificity of 98% for the probable hyperactivity
disorder category. The positive predictive power was also fairly
low at 27%. Examples of scores that exceed the threshold for this
model using optimal values are given in the supporting informa-
tion in Table S1.
LR was also used to explore the predictive value of the subscales
for ASD (Table 2).
Again all the SDQ subscales were significant in unadjusted
analysis. After adjustment for interdependencies between sub-
scales, several still remained significantly associated with the
outcome of ASD at 10% levels. These were the impact and
hyperactivity subscales from both raters, and the prosocial and
emotional symptoms scores rated by parents. The measures with
the largest effect were the parent-rated subscales of the prosocial
behaviour and impact subscales. Peer problems from either rater
did not appear in the final model. A threshold of 0.03 produced
the optimal values for model sensitivity and specificity of 79% and
93% respectively; AUC = 0.90, (95% CI, 0.860.95). The PPV
was again low at 18%. Examples of scores that exceed the
threshold for this ASD model using optimal values are given in the
supporting information in Table S2. Table 3 shows the threshold,
sensitivity and specificity for higher PPVs for both the ASD and
ADHD model, illustrating the varying sensitivity, specificity and
predictive power of the model at various threshold settings.
Discussion
The prevalence of ASD and ADHD was not the focus of this
paper: we have written about this elsewhere [44]. The low
prevalence of parent-reported ADHD diagnosis is consistent with
other UK studies [45] and studies in Scandinavia [46]. The
reported prevalence of ASD diagnosis is high compared to
previous estimates; which may reflect the increasing use of the
ASD label in the UK, a trend that has also been identified in other
studies. Results showed elevated behavioral difficulties in multiple
domains for both groups with parent-reported diagnoses, and
suggests that many behavioral problems are shared by children
diagnosed with ASD and those diagnosed with ADHD.
Despite the exclusivity clause in the current ICD-10 diagnostic
classification systems, there was a high proportion of dual
diagnosis in the two conditions: 23% of children with ADHD
had a diagnosis of ASD, and 21% with ASD had identified
ADHD. Several other recent studies [2933] also suggest that
children with ASD and ADHD often share symptoms of
5 December 2013 | Volume 8 | Issue 12 | e80247
 The Association of SDQ with Diagnosis of ASD/ADHD

Table 2. Models of SDQ sub-scales as predictors with ASD diagnosis as outcome for children from Milenium Cohort at age 7.

Unadjusted Unadjusted Unadjusted Adjusted Adjusted OR Adjusted Final ASD OR

Variable n OR & 95% CI p n & 95% CI p1 Final n & 95% CI3 Final p

Emotion 13127 1.52 (1.41,1.64) ,0.001 8162 1.15 (1.00,1.33) 0.04 8180 1.16 (1.01,1.33) 0.04
Parent
Emotion 8536 1.38 (1.29,1.48) ,0.001 8162 1.05 (0.93,1.20) 0.42
Teacher
Conduct 13155 1.63 (1.53,1.74) ,0.001 8162 0.92 (0.80,1.06) 0.26
Parent
Conduct 8539 1.41 (1.33,1.50) ,0.001 8162 0.83 (0.74,0.95) 0.005 8180 0.86 (0.76,0.96) 0.01
Teacher
Hyper 13105 1.75 (1.63,1.88) ,0.001 8162 1.18 (1.04,1.33) 0.009 8180 1.15 (1.03,1.30) 0.02
Parent
Hyper 8537 1.48 (1.38,1.58) ,0.001 8162 1.10 (1.00,1.22) 0.06 8180 1.11 (1.01,1.23) 0.03
Teacher
Prosocial 13159 0.54 (0.49,0.59) ,0.001 8162 1.24 (1.10,1.41) 0.001 8180 1.25 (1.11,1.42) ,0.001
Parent2
Prosocial 8535 0.64 (0.59,0.70) ,0.001 8162 1.10 (0.97,1.24) 0.14
Teacher2
Peer 13136 1.94 (1.80,2.09) ,0.001 8162 1.04 (0.88,1.23) 0.62
Parent
Peer 8536 1.65 (1.53,1.77) ,0.001 8162 1.06 (0.91,1.23) 0.46
Teacher
Impact 13004 2.11 (1.93,2.30) ,0.001 8162 1.51 (1.30,1.75) ,0.001 8180 1.53 (1.35,1.72) ,0.001
Parent
Impact 8430 2.31 (2.10,2.54) ,0.001 8162 1.25 (1.02,1.55) 0.04 8180 1.44 (1.22,1.71) ,0.001
Teacher

1 Adjusted models include subscales significant at 10% levels. 2.For prosocial scores the reciprocal of the odds ratios.
Is presented to fit conceptually with the rest of the model, i.e. greater OR = greater association with ASD. 3. Model constant 0.113 on odds ratio scale.
doi:10.1371/journal.pone.0080247.t002

hyperactivity and other behavioural difficulties. ADHD symptoms
are relatively common in children and adults with autistic-type
symptoms; autism-type symptoms/ behaviours may be less
common in children with ADHD [27,28,30,31]. Our findings of
elevated behavioral difficulties indicative of both conditions in
both diagnosed groups support change to the diagnostic criteria to
allow ASD and ADHD to be diagnosed in the same individual.
Our findings suggest that this already relatively common in
practice, so removal of the exclusivity clauses would eliminate
unnecessary tension between clinical practice and diagnostic rules.
After adjustment for other subscales in multivariable models, the
final model for ADHD was composed of the hyperactivity/
inattention and impact symptoms only. This finding is highly
predictable and as initially hypothesised, although the selection
biases inherent in obtaining a clinical diagnosis may have clouded
the relationship. Although the findings suggest that ADHD
symptoms are also relatively common in children with ASD and
vice-versa, in line with findings from other studies [2733], the
results do not support the argument that ASD and ADHD should
be considered as different manifestations of one overarching
disorder [33,34].
In our study LR models, after statistical adjustment for
interdependencies between different types of behavioral problems,
a distinctive symptom profile emerged for ADHD based on
hyperactivity and impact sub-scales, but not for ASD. The finding
provides evidence to support the assertion of Nicalsen et al. [22]
that the SDQ hyperactivity-inattentive subscale shows good
agreement with the diagnostic criteria for attention-deficit

Table 3. Varying sensitivity, specificity and positive predictive values for ASD and ADHD models derived from Millennium Cohort
Data.

PPV ADHD ASD

Threshold Sensitivity Specificity Threshold Sensitivity Specificity

20% 0.05 81% 95% 0.04 76% 94%

30% 0.13 63% 98% 0.08 66% 97%
40% 0.39 34% 99% 0.17 53% 98%
50% 0.74 15% 100% 0.28 47% 99%
60% 0.92 5% 100% 0.57 29% 100%

doi:10.1371/journal.pone.0080247.t003

PLOS ONE | www.plosone.org 6 December 2013 | Volume 8 | Issue 12 | e80247


hyperactivity disorder, as it was intended to do. Our adjusted
results suggest that children with ADHD have focused problems of
hyperactivity/inattention. A combined model derived from both
parent and teacher hyperactivity and impact scores is a good
predictor of diagnosis of ADHD, identifying up to 91% of children
with parent-reported clinical diagnosis of ADHD. The models
developed using the MCS data and the Goodman PHD algorithm
differ in two ways. First, the MCS model did not find impact on
teachers to be a significant predictor of ADHD. This contradicts
arguments of many socially orientated researchers who suggest
that ADHD is partially constructed in response to the need for
compliance at school [47]. Others have observed that ADHD is
more likely to be identified in tandem with disruption to the
classroom [48]. MCS data suggests for teachers, presence of
inattention and hyperactivity alone is enough to indicate ADHD.
One partial explanation could be that naming the condition:
ADHD being diagnosed; minimises teacher ratings of impact.
The second main difference is that cut-offs (e.g. for identifying
91% of children with disorder) were not fixed as are those in the
PHD model. This is consistent with the findings of Ullebo and
colleagues [20], who conclude, and the ROC curves demonstrate
that thresholds can be selected by defining a specificity or
sensitivity value to obtain specified model performance. Appro-
priate cut-off can then be chosen according to purpose of use. The
coefficients for the logistic regression models can be obtained from
the odds ratios in Tables 1 and 2. In a clinical setting, the
probability of an ADHD/ASD diagnosis can be calculated given a
set of SDQ scores. The probability of a diagnosis can then be
compared to the optimal threshold.
Goodman and Mullick [12] and Ullebo and colleagues [20]
cautiously recommend use of the SDQ as a screening tool for
childhood disorder and specifically ADHD/hyperkinetic disor-
der, Brndbo and colleagues [25] caution against it. All these
studies used well-validated scales measuring symptoms of ADHD.
Our study used an outcome measure of parent-reported clinical
diagnosis of disorder: as clinical assessments are highly variable
and subject to local bias [49], our findings have no clinical
application until replicated against standardised ADHD scales. It
should be remembered that previous work on the algorithm [21]
predicted against diagnoses made using a research instrument,
while the current study uses parent-report of a clinical diagnosis;
both studies report from a general population sample. For MCS,
the PHD algorithm had low sensitivity at 30%, but a specificity
of 98%.
The resulting LR model for autism shows that many types of
difficulties may complicate the picture for a child with ASD. This
is to be expected, as there is not a specific autism spectrum
subscale that focuses on the core difficulties as there is with
ADHD. Prosocial behaviour emerged as the strongest predictor of
ASD, which again is not surprisingly as social impairments are
core deficits. Furthermore, ASD diagnosis has been associated
with the low scores on the prosocial subscale in other UK cohorts
[50]. Our findings suggest that a range of other difficulties such as
anxiety and conduct problems are likely to commonly co-occur
with both ASD and ADHD, which, for those working with
children who have these difficulties, echoes clinical experience. It is
intriguing that ASD is not associated with conduct problems;
indeed higher conduct problem ratings lower the odds of an ASD
diagnosis. It may be that social difficulties inhibit the overt
externalising behaviours covered by the SDQ, several of which
require a social orientation towards others. Behaviour that
challenges others among children with ASD often results from a
failure to recognise or conform to social expectations and/or
rigidity around routine or preferred activity, which may not be
PLOS ONE | www.plosone.org
The Association of SDQ with Diagnosis of ASD/ADHD
adequately tapped by the SDQ behaviour subscale. ASD was
associated with enhanced emotional problems. These results
concur with many studies that have found ASD to be associated
with anxiety and depression [5153]. Taking account of co-
occurring symptoms is essential for any child with autism as it may
have practical ramifications in terms of the type(s) of intervention
required.
Limitations
The current study used parent-report of clinical identification of
ASD and ADHD by a doctor or another professional. This means
that parents are likely to be well aware of symptoms of these
conditions and may therefore be more likely to report them than a
parent of a child with similar difficulties that have not been
clinically highlighted. Furthermore, parents may have been over-
inclusive in their interpretation of the question: inferring a positive
answer in cases where ASD or ADHD was suggested by a health
worker but not confirmed by further assessment. Clinician
diagnoses themselves can be inaccurate if unguided by structured
assessment [54].
In addition, the sample will contain other children with ASD
and ADHD, and other disorders, as yet unrecognised [55], and
research suggests that the unrecognised group may be in the
majority [56]. It is beyond the scope of this article to comment on
differentiation from other comorbid groups. As children with other
disorders were mixed in with general population, the mean SDQ
scores of the general population are likely to have been elevated
and would serve to make the mean SDQ scores more similar to
those of children with parent- reported ASD and ADHD. Thus,
our detection of differences between children with a clinical
diagnosis of ADHD and ASD is likely to be robust.
Other limitations relate to the ADHD group. First, we did not
have access to pharmaceutical data, but evidence suggests
treatment with methylphenidate may have improved symptoms
of hyperactivity and may have led parents and teachers to under-
report difficulties in any children with diagnoses who were taking
medication [57]. Second, we did not have information on sub-type
of ADHD. Ullebo and colleagues [20] found the SDQ was a good
predictor for the combined ADHD subtype, but less informative
for other subtypes. Third, seven years old is still early in life for
clinical identification of ADHD [58], which may partially explain
why the sensitivity of the PHD algorithm was so low in our study.
It identified less than a third of the children with ADHD, a much
poorer performance than witnessed in some other studies [20,21].
At this age it is likely many cases are yet to be identified:
assessment with a research-based diagnostic measure may have
revealed different results.
Despite these limitations our results provide further evidence to
suggest that the SDQ algorithm is a useful tool as an indicator of
ADHD symptoms for research purposes. As the SDQ instrument
is widely used in research studies already [613], bespoke cut-offs
could be developed according to purpose of application to
research. However, we do not currently recommend using the
SDQ as a screening tool for either disorder in clinical practice due
to the high number of false positives and limitations of case
definition in our study. Should our findings be replicated against
structured research assessments, they could be used by clinicians to
identify children at risk of ADHD who warrant further assessment.
The study is part of a large and growing literature that
demonstrates that ADHD symptoms are also relatively common in
children and adults diagnosed with ASD and vice-versa [2733]. It
supports changes to DSM-5 dropping the exclusivity clause to
allow dual diagnosis of ASD and ADHD, and suggests ICD
criteria should follow suit: indeed dual diagnosis occurs in practice
7 December 2013 | Volume 8 | Issue 12 | e80247

already. These results also suggest that for children with ASD, the
presence of other co-occurring impairments in behaviour is likely
to be the rule, not the exception. In this way the work contributes
to the debate raised by Hattori et al. [33,34] about whether the
current diagnostic configuration for ADHD and autism are valid.
The findings support the removal of exclusivity clauses in current
revisions to DSM-5, and inform on-going debates about revisions
to ICD-11.
Supporting Information
Table S1 Examples of scores over the threshold for
ADHD model. There are 1331 combinations of 3 SDQ scales
(010) of which 928 combinations would produce a value over the
threshold. The most frequent combinations in MCS are included
for illustrative purposes.
(DOCX)
References
1. Goodman R (1997) The Strengths and Difficulties Questionnaire: a research
note. J Child Psychol Psychiatry 38: 5816.
2. Goodman R (1999) The extended version of the Strengths and Difficulties
Questionnaire as a guide to child psychiatric caseness and consequent burden.
J Child Psychol Psychiatry 40: 7919.
3. Levitt JM, Saka N, Romanelli LH, Hoagwood K (2007) Early identification of
mental health problems in schools: The status of instrumentation. J Sch Psychol
45: 16391.
4. Borg A-M, Kaukonen P, Salmelin R, Joukamaa M, Tamminen T (2012)
Reliability of the Strengths and Difficulties Questionnaire among Finnish 49-
year-old children. Nord J Psychiatry 66: 40313.
5. Yao S, Zhang C, Zhu X, Jing X, McWhinnie CM, et al. (2009) Measuring
adolescent psychopathology: psychometric properties of the self-report Strengths
and Difficulties Questionnaire in a sample of Chinese adolescents. J. Adolesc
Health 56: 5562.
6. Muris P, Meesters C, van den Berg F (2012) The Strengths and Difficulties
Questionnaire (SDQ). Eur Child Adolesc Psychiatry 12: 18.
7. Obel C, Heiervang E, Rodriguez A, Heyerdahl S, Smedje H, et al. (2004) The
Strengths and Difficulties Questionnaire in the Nordic countries. Eur Child
Adolesc Psychiatry 13 Suppl 2: II3239.
8. Woerner W, Fleitlich-Bilyk B, Martinussen R, Fletcher J, Cucchiaro G, et al.
(2004) The Strengths and Difficulties Questionnaire overseas: evaluations and
applications of the SDQ beyond Europe. Eur. Child Adolesc Psychiatry 13: 47
54.
9. Dickey WC, Blumberg SJ (2001) Revisiting the factor structure of the strengths
and difficulties questionnaire: United States. J Am Acad Child Adolesc
Psychiatry 43:115967.
10. Mason WA, Chmelka MB, Thompson RW (2012) Responsiveness of the
Strengths and Difficulties Questionnaire (SDQ) in a sample of high-risk youth in
residential treatment. Child Youth Care Forum. 41: 47992.
11. Leung PWL, Wong MMT (2003) Measures of child and adolescent
psychopathology in Asia. Psychol Assess 15: 26879.
12. Mullick MSI, Goodman R (2001) Questionnaire screening for mental health
problems in Bangladeshi children: a preliminary study. Soc. Psychiatry Psychiatr
Epidemiol 36: 9499.
13. Kashala E, Elgen I, Sommerfelt K, Tylleskar T (2005) Teacher ratings of mental
health among school children in Kinshasa, Democratic Republic of Congo. Eur
Child Adolesc Psychiatry 14: 20815.
14. Chittleborough CR, Lawlor DA, Lynch JW. (2011) Young maternal age and
poor child development: predictive validity from a birth cohort. Pediatrics. 127:
E1436E1444.
15. Griffiths LJ, Dezateux C, Hill A (2011) Is obesity associated with emotional and
behavioural problems in children? Findings from the Millennium Cohort Study.
Int. J. Pediatr Obes 22: E423E432.
16. Stone LL, Otten R, Engels RCME, Vermulst AA, Janssens JMAM (2010)
Psychometric properties of the parent and teacher versions of the Strengths and
Difficulties Questionnaire for 4-to 12-year-olds: a review. Clin Child Fam
Psychol Rev 3: 25474.
17. Collishaw S, Maughan B, Goodman R, Pickles A (2004) Time trends in
adolescent mental health. J Child Psychol Psychiatry 45: 135062.
18. Goodman R, Renfrew D, Mullick M (2000) Predicting type of psychiatric
disorder from Strengths and Difficulties Questionnaire (SDQ) scores in child
mental health clinics in London and Dhaka. Eur. Child Adolesc Psychiatry 9:
12934.
19. Mathai J, Anderson P, Bourne A (2004) Comparing psychiatric diagnoses
generated by the Strengths and Difficulties Questionnaire with diagnoses made
by clinicians. Aust N Z J Psychiatry 38: 63943.
PLOS ONE | www.plosone.org
The Association of SDQ with Diagnosis of ASD/ADHD
Table S2 Examples of scores over the threshold for ASD
model. In MCS data there are 671 combinations which produce
a value over the threshold. They are all unique. Table S2 gives 10
combinations which are over the threshold, for illustrative
purposes.
(DOCX)
Acknowledgments
We would like to thank the Millennium Cohort Study families for their
time and cooperation, as well as the Millennium Cohort Study team at the
Institute of Education, London, UK. We would also like to thank Dr.
Obioha Ukoumunne for his helpful comments on this article.
Author Contributions
Conceived and designed the experiments: GR LR TF. Performed the
experiments: LR GR. Analyzed the data: GR LR. Wrote the paper: GR
LR TF.
20. Ullebo AK, Posserud M-B, Heiervang E, Gillberg C, Obel C (2011) Screening
for the attention deficit hyperactivity disorder phenotype using the Strength and
Difficulties Questionnaire. Eur Child Adolesc Psych 20(9): 4518.
21. Goodman R, Ford T, Simmons H, Gatward R, Meltzer H (2000) Using the
Strengths and Difficulties Questionnaire (SDQ) to screen for child psychiatric
disorders in a community sample. BJP. 177(6): 5349.
22. Niclasen J, Teasdale TW, Andersen A-MN, Skovgaard AM, Elberling H, et al.
(2012) Psychometric Properties of the Danish Strength and Difficulties
Questionnaire: the SDQ assessed for more than 70,000 raters in four different
cohorts. PLoS One 7(2). Available: http://www.ncbi.nlm.nih.gov/pmc/articles/
PMC3288071 Accessed 2013 Jan 25.
23. World Health Organization. International classification of diseases and related
health problems (ICD-10). [Internet]. Geneva, WHO; 1992. Available: www.
who.int/classifications/icd/en/greenbook.pdf p.179 Accessed 2013 Jan 25.
24. American Psychiatric Association, DSM-5 Task Force (2013) Diagnostic and
statistical manual of mental disorders: DSM-5. Arlington, Va.: American
Psychiatric Association.
25. Brndbo PH, Mathiassen B, Martinussen M, Heiervang E, Eriksen M, et al.
(2011) The Strengths and Difficulties Questionnaire as a screening instrument
for Norwegian child and adolescent mental health services, application of UK
scoring algorithms. Child Adolesc Psychiatry 5(1): 32.
26. Iizuka C, Yamashita Y, Nagamitsu S, Yamashita T, Araki Y, et al. (2010)
Comparison of the Strengths and Difficulties Questionnaire (SDQ) scores
between children with high-functioning autism spectrum disorder (HFASD) and
attention-deficit/hyperactivity disorder (AD/HD). Brain Dev 32: 60912.
27. Hofvander B, Delorme R, Chaste P, Nyden A, Wentz E, et al. (2009) Psychiatric
and psychosocial problems in adults with normal-intelligence autism spectrum
disorders. BMC Psychiatry 9: 35.
28. Reiersen AM, Constantino JN, Grimmer M, Martin NG, Todd RD (2008)
Evidence for shared genetic influences on self-reported ADHD and autistic
symptoms in young adult Australian twins. Twin Res Hum Genet 11(6): 57985.
29. Reiersen AM, Todd RD (2008) Co-occurrence of ADHD and autism spectrum
disorders: phenomenology and treatment. Expert Rev Neurother 8(4): 65769.
30. Santosh PJ, Mijovic A (2004) Social impairment in Hyperkinetic Disorder
relationship to psychopathology and environmental stressors. Eur Child Adolesc
Psychiatry. 3: 14150.
31. Simonoff E, Pickles A, Charman T, Chandler S, Loucas T, et al. (2008)
Psychiatric disorders in children with autism spectrum disorders: prevalence,
comorbidity, and associated factors in a population-derived sample. J Am Acad
Child Adolesc Psychiatry 8: 9219.
32. St Pourcain B, Mandy WP, Heron J, Golding J, Davey Smith G, et al. (2011)
Links between co-occurring social-communication and hyperactive-inattentive
trait trajectories. J Am Acad Child Adolesc Psychiatry 50(9): 892902.e5.
33. Van der Meer JMJ, Oerlemans AM, van Steijn DJ, Lappenschaar MGA, de
Sonneville LMJ, et al. (2012) Are autism spectrum disorder and attention-
deficit/hyperactivity disorder different manifestations of one overarching
disorder? Cognitive and symptom evidence from a clinical and population-
based sample. J Am Acad Child Adolesc Psychiatry 51: 11601172.e3.
34. Hattori J, Ogino T, Abiru K, Nakano K, Oka M, et al. (2006) Are pervasive
developmental disorders and attention-deficit/hyperactivity disorder distinct
disorders? Brain Dev 28(6): 3714.
35. University of London. Institute of Education. Centre for Longitudinal Studies
(2008) Millennium Cohort Study: Fourth Survey, [computer file]. 4th Edition.
Colchester, Essex: UK Data Archive [distributor], SN: 6411, http://dx.doi.org/
10.5255/UKDA-SN-6411-3.
36. Shepherd P (2012). Millennium Cohort Study: Ethical review and consent.
Centre for Longitudinal Studies, London: UK.
8 December 2013 | Volume 8 | Issue 12 | e80247

37. Plewis I (2007) The Millennium Cohort study: technical report on sampling.
Centre for Longitudinal Studies, London; UK. Available: http://www.esds.ac.
uk/doc/5350/mrdoc/pdf/mcs_technical_report_on_sampling_4th_edition.pdf
Accessed 2013 Jan 25.
38. Pastor PN, Rueben CA (2008) Diagnosed attention deficit hyperactivity disorder
and learning disability: United States, 20042006. US Department of Health
and Human Services, Centres for disease control and prevention, Office of
Analysis and Epidemiology. Vital and health statistics series 10, no. 237.
39. Totsika V, Hastings R, Emerson E, Berridge D, Lancaster G (2011) Behavior
problems at 5 years of age and maternal mental health in autism and intellectual
disability. J Ab Child Psychol 39: 113747.
40. Boyle CA, Boulet S, Schieve LA, Cohen RA, Blumberg SJ, et al. (2011) Trends
in the prevalence of developmental disabilities in US children, 19972008.
Pediatrics. 127(6): 103442.
41. Kogan MD, Blumberg SJ, Schieve LA, Boyle CA, Perrin JM, et al. (2007)
Prevalence of parent-reported diagnosis of autism spectrum disorder among
children in the US Pediatrics 124: 1395403.
42. Goodman R, Ford T, Corbin T, Meltzer H (2004) Using the Strengths and
Difficulties Questionnaire (SDQ) multi-informant algorithm to screen looked-
after children for psychiatric disorders. Eur Child Adolesc Psychiatry Suppl 2:
II2531.
43. Fluss R, Faraggi D, Reiser B (2005) Estimation of the Youden Index and its
associated cutoff point. Biom J. Aug;47: 45872.
44. Russell G, Rodgers LR, Ukoumunne OC, Ford T (2013) Prevalence of parent-
reported ASD and ADHD in the UK: findings from the Millennium Cohort
Study. J Autism Dev Disord. (e pub ahead of print).
45. Ford T, Goodman R, Meltzer H (2003) The British Child and Adolescent
Mental Health Survey 1999: the prevalence of DSM-IV disorders. J Am Acad
Child Adolesc Psychiatry. Oct;42: 120311.
46. Kvist AP, Nielsen HS, Simonsen M (2013) The importance of childrens ADHD
for parents relationship stability and labor supply. Soc Sci Med. 88: 3038.
PLOS ONE | www.plosone.org
The Association of SDQ with Diagnosis of ASD/ADHD
47. Bailey S, Thomson P (2009) Routine (dis)order in an infant school. Ethnogr
Educ. 4: 21127.
48. Abikoff HB, Jensen PS, Arnold LLE, Hoza B, Hechtman L, et al. (2002)
Observed classroom behavior of children with ADHD: relationship to gender
and comorbidity. J Abnorm Child Psychol. 1;30: 34959.
49. Reid R, Hakendorf P, Prosser B (2002) Use of psychostimulant medication for
ADHD in South Australia. J Am Acad Child Adolesc Psychiatry 41: 90613.
50. Russell G, Steer C, Golding J (2011) Social and demographic factors that
influence the diagnosis of autistic spectrum disorders. Soc Psychiatry Psychiatr
Epidemiol 46: 128393.
51. Hallett V, Ronald A, Happe F (2009) Investigating the association between
autistic-like and internalizing traits in a community-based twin sample. J Am
Acad Child Adolesc Psychiatr 48: 61827.
52. Smalley SL, McCracken J, Tanguay P (1995) Autism, affective disorders, and
social phobia. Am J Med Gen 60: 1926.
53. Tantam D (2000) Psychological disorder in adolescents and adults with Asperger
syndrome. Autism 4(1): 4762.
54. Bruchmuller K, Margraf J, Schneider S (2012) Is ADHD diagnosed in accord
with diagnostic criteria? Overdiagnosis and influence of client gender on
diagnosis. J Consult Clin Psychol 80: 12838.
55. Meltzer H, Gatward R, Goodman R, Ford T (2003) Mental health of children
and adolescents in Great Britain. Int Rev Psychiatry 15: 1857.
56. Ford T, Hamilton H, Meltzer H, Goodman R (2007) Child mental health is
everybodys business: the prevalence of contact with public sector services by
type of disorder among British school children in a three-year period. Child
Adolesc Ment Health 12(1): 13.
57. Barbaresi WJ, Katusic SK, Colligan RC, Weaver AL, Leibson CL, et al. (2006)
Long-term stimulant medication treatment of attention-deficit/hyperactivity
disorder: results from a population-based study. J Dev Behav Pediatr. 27: 110.
58. Kieling C, Kieling RR, Rohde LA, Frick PJ, Moffitt T, et al. (2010) The age at
onset of attention deficit hyperactivity disorder. Am J Psychiatry 167: 146.
9 December 2013 | Volume 8 | Issue 12 | e80247